Re: Request for comments - Keras-MXNet as submodule in MXNet

[sandeep krishnamurthy]

1. I agree bringing whole keras-mxnet repo as submodule in MXNet is not the
right thing to do. I was mainly proposing, a stripped version of
keras-mxnet which has keras interface with MXNet backend in to MXNet
enabling MXNet users to use Keras interface natively within MXNet (Ex:
mx.keras)

2. Review process in keras-team/keras has not passed due to design
incompatibility with MXNet - Keras and NOT due to any Keras API interface
issue or functionality exposed to users.

Let me rephrase my proposal for comments - "Do we want to bring in Keras
interface into MXNet?"

Best,
Sandeep

On Fri, Mar 23, 2018 at 12:06 PM, Xingjian SHI 
wrote:

> It should at least pass the review process. However, the Keras-MXNet has
> not passed the review by the keras team.
>
>
> Best,
>
> Xingjian
>
>
> 
> From: Naveen Swamy 
> Sent: Saturday, March 24, 2018 3:04 AM
> To: dev@mxnet.incubator.apache.org
> Subject: Re: Request for comments - Keras-MXNet as submodule in MXNet
>
> define - not mature?
>
> Are all new commits that go into MXNet mature ?
>
> On Fri, Mar 23, 2018 at 12:02 PM, Xingjian SHI 
> wrote:
>
> > -1. We do not need to get the eyeballs this way. Asking users to use
> > features that are not mature enough will only hurt the future development
> > of MXNet.
> >
> >
> > Xingjian
> >
> > 
> > From: Naveen Swamy 
> > Sent: Saturday, March 24, 2018 2:54 AM
> > To: dev@mxnet.incubator.apache.org
> > Subject: Re: Request for comments - Keras-MXNet as submodule in MXNet
> >
> > The proposal is about bringing a forked version of Keras(that works only
> > with MXNet) into Apache MXNet repo submodule that way MXNet gets more
> > eyeballs from existing Keras users and eventually Gluon, etc., , like
> > Sandeep mentioned Keras has a large user base which MXNet could tap into.
> >
> > On Fri, Mar 23, 2018 at 11:50 AM, Yao Wang 
> > wrote:
> >
> > > -1 Creating Keras as submodule of MXNet will provide users a feeling
> that
> > > MXNet depends on Keras. Keras is a frontend library which can be
> > supported
> > > by various different backend framework. It would be better to add
> backend
> > > framework as Keras's submodule(Keras depends on MXNet) rather than
> > > opposite.
> > >
> > > Best,
> > > Yao
> > >
> > > 2018-03-23 11:44 GMT-07:00 Xingjian SHI :
> > >
> > > > -1. I think we should wait until it's merged into keras-team/keras.
> The
> > > > repo is still not mature enough.
> > > >
> > > >
> > > > Best,
> > > >
> > > > Xingjian
> > > >
> > > >
> > > > 
> > > > From: sandeep krishnamurthy 
> > > > Sent: Friday, March 23, 2018 1:49 PM
> > > > To: dev@mxnet.incubator.apache.org
> > > > Subject: Request for comments - Keras-MXNet as submodule in MXNet
> > > >
> > > > Hello MXNet Community,
> > > >
> > > > Along with Lai, Karan and other MXNet contributors, I am working on
> > > adding
> > > > MXNet backend for Keras. Currently supporting around ~70% of Keras
> APIs
> > > > across CNNs and RNNs.
> > > > https://github.com/deep-learning-tools/keras/tree/
> keras2_mxnet_backend
> [https://avatars0.githubusercontent.com/u/32447491?s=400=4] github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
>
> deep-learning-tools/keras tools/keras/tree/keras2_mxnet_backend>
> github.com
> keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or
> CNTK.
>
>
>
> > [https://avatars0.githubusercontent.com/u/32447491?s=400=4] > github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
> >
> > deep-learning-tools/keras > tools/keras/tree/keras2_mxnet_backend>
> > github.com
> > keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or
> > CNTK.
> >
> >
> >
> > > > [https://avatars0.githubusercontent.com/u/32447491?s=400=4
> ] > > > github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
> > > >
> > > > deep-learning-tools/keras > > > tools/keras/tree/keras2_mxnet_backend>
> > > > github.com
> > > > keras - Deep Learning library for Python. Runs on TensorFlow, Theano,
> > or
> > > > CNTK.
> > > >
> > > >
> > > >
> > > >
> > > > We wanted to gather the community feedback on the proposal for
> > including
> > > > this keras-mxnet package as a submodule in Apache MXNet. This will
> > enable
> > > > providing the Keras interface for MXNet users. MXNet users can choose
> > > Keras
> > > > interface for building their Neural Networks in Symbolic Mode (Ex:
> > > > mx.keras).
> > > >
> > > > *Advantages:*
> > > >
> > > > 1. Keras is widely popular interface that many DL practitioners are
> > > > familiar. By including keras interface within MXNet natively, we
> enable
> > > > many users to use MXNet with 0 

Re: MXNet C++ package improvements

[Tianqi Chen]

One quote that I really liked and related to this topic “the best design is
achieved not when you have nothing to add, but when there is nothing to be
taken away”.

Most of the stable api decision we put into a project will become a
technical debt, and a minimum clean design with equally powerful api adds
more value than a bigger one
On Fri, Mar 23, 2018 at 12:11 PM Tianqi Chen 
wrote:

> +1 for hour-glass CAPI design. It is there for very good reason. For
> example, if you are windows user an want to build your app using mingw,
> current way works. While the raw c++ approach won’t due to the fact cuda is
> only supported by MSVC.
>
> The argument that c++ api itself is more elegant and fast is not
> necessarily valid. Always remember Amdahl’s Law, the additional overhead
> saved by directly using are not the bottleneck.
>
> The effectiveness of core engine, on the other hand will be on the
> critical path, and we rely on constantly improving this which can mean
> break backward compatibility if C++ api layer is exposed
>
> Tianqi
>
>
> On Fri, Mar 23, 2018 at 11:45 AM Eric Xie  wrote:
>
>> I see several issues with the design. I've commented in the document but
>> for record here:
>>
>> 1. cpp-package is almost only used for inference. since you are planning
>> a rewrite that's almost certainly non-backward-compatible, we might as well
>> create a new interface that's inference only.
>>
>> 2. The hour-glass CAPI design should be kept.
>>
>> On 2018/03/14 18:07:49, Anton Chernov  wrote:
>> > Dear MxNet Community,
>> >
>> > please find here
>> > <
>> https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=75976112>
>> the
>> > design document for the proposed MXNet C++ package improvements for
>> review
>> > and consideration.
>> >
>> > Feedback is welcome and highly appreciated. Thank you!
>> >
>> > BR
>> > Anton
>> >
>>
>

Re: MXNet C++ package improvements

[Tianqi Chen]

+1 for hour-glass CAPI design. It is there for very good reason. For
example, if you are windows user an want to build your app using mingw,
current way works. While the raw c++ approach won’t due to the fact cuda is
only supported by MSVC.

The argument that c++ api itself is more elegant and fast is not
necessarily valid. Always remember Amdahl’s Law, the additional overhead
saved by directly using are not the bottleneck.

The effectiveness of core engine, on the other hand will be on the critical
path, and we rely on constantly improving this which can mean break
backward compatibility if C++ api layer is exposed

Tianqi


On Fri, Mar 23, 2018 at 11:45 AM Eric Xie  wrote:

> I see several issues with the design. I've commented in the document but
> for record here:
>
> 1. cpp-package is almost only used for inference. since you are planning a
> rewrite that's almost certainly non-backward-compatible, we might as well
> create a new interface that's inference only.
>
> 2. The hour-glass CAPI design should be kept.
>
> On 2018/03/14 18:07:49, Anton Chernov  wrote:
> > Dear MxNet Community,
> >
> > please find here
> > <
> https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=75976112>
> the
> > design document for the proposed MXNet C++ package improvements for
> review
> > and consideration.
> >
> > Feedback is welcome and highly appreciated. Thank you!
> >
> > BR
> > Anton
> >
>

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Xingjian SHI]

It should at least pass the review process. However, the Keras-MXNet has not 
passed the review by the keras team.


Best,

Xingjian



From: Naveen Swamy 
Sent: Saturday, March 24, 2018 3:04 AM
To: dev@mxnet.incubator.apache.org
Subject: Re: Request for comments - Keras-MXNet as submodule in MXNet

define - not mature?

Are all new commits that go into MXNet mature ?

On Fri, Mar 23, 2018 at 12:02 PM, Xingjian SHI 
wrote:

> -1. We do not need to get the eyeballs this way. Asking users to use
> features that are not mature enough will only hurt the future development
> of MXNet.
>
>
> Xingjian
>
> 
> From: Naveen Swamy 
> Sent: Saturday, March 24, 2018 2:54 AM
> To: dev@mxnet.incubator.apache.org
> Subject: Re: Request for comments - Keras-MXNet as submodule in MXNet
>
> The proposal is about bringing a forked version of Keras(that works only
> with MXNet) into Apache MXNet repo submodule that way MXNet gets more
> eyeballs from existing Keras users and eventually Gluon, etc., , like
> Sandeep mentioned Keras has a large user base which MXNet could tap into.
>
> On Fri, Mar 23, 2018 at 11:50 AM, Yao Wang 
> wrote:
>
> > -1 Creating Keras as submodule of MXNet will provide users a feeling that
> > MXNet depends on Keras. Keras is a frontend library which can be
> supported
> > by various different backend framework. It would be better to add backend
> > framework as Keras's submodule(Keras depends on MXNet) rather than
> > opposite.
> >
> > Best,
> > Yao
> >
> > 2018-03-23 11:44 GMT-07:00 Xingjian SHI :
> >
> > > -1. I think we should wait until it's merged into keras-team/keras. The
> > > repo is still not mature enough.
> > >
> > >
> > > Best,
> > >
> > > Xingjian
> > >
> > >
> > > 
> > > From: sandeep krishnamurthy 
> > > Sent: Friday, March 23, 2018 1:49 PM
> > > To: dev@mxnet.incubator.apache.org
> > > Subject: Request for comments - Keras-MXNet as submodule in MXNet
> > >
> > > Hello MXNet Community,
> > >
> > > Along with Lai, Karan and other MXNet contributors, I am working on
> > adding
> > > MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> > > across CNNs and RNNs.
> > > https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
[https://avatars0.githubusercontent.com/u/32447491?s=400=4]

deep-learning-tools/keras
github.com
keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or CNTK.



> [https://avatars0.githubusercontent.com/u/32447491?s=400=4] github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
>
> deep-learning-tools/keras tools/keras/tree/keras2_mxnet_backend>
> github.com
> keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or
> CNTK.
>
>
>
> > > [https://avatars0.githubusercontent.com/u/32447491?s=400=4] > > github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
> > >
> > > deep-learning-tools/keras > > tools/keras/tree/keras2_mxnet_backend>
> > > github.com
> > > keras - Deep Learning library for Python. Runs on TensorFlow, Theano,
> or
> > > CNTK.
> > >
> > >
> > >
> > >
> > > We wanted to gather the community feedback on the proposal for
> including
> > > this keras-mxnet package as a submodule in Apache MXNet. This will
> enable
> > > providing the Keras interface for MXNet users. MXNet users can choose
> > Keras
> > > interface for building their Neural Networks in Symbolic Mode (Ex:
> > > mx.keras).
> > >
> > > *Advantages:*
> > >
> > > 1. Keras is widely popular interface that many DL practitioners are
> > > familiar. By including keras interface within MXNet natively, we enable
> > > many users to use MXNet with 0 learning curve.
> > >
> > > 2.  Adding as submodule and exposing natively within MXNet pip package,
> > > would greatly enhance user experience and get more users as compared to
> > > releasing a fork repository independently.
> > >
> > > 3. Why submodule? - Helps in easily managing with patching the latest
> > > parent keras-team/keras developments and releases. Thereby helping us
> > > provide users the core keras experience. Operational management.
> > >
> > > 4. Other minor advantages - Operational maintenance, pip, CI and
> quality
> > > control.
> > >
> > > Please do share your comments on the proposal.
> > >
> > > Best,
> > > Sandeep
> > >
> > > *Note: *We tried merging with keras-team/keras and we created a PR
> > >  as well. However, due
> to
> > > multiple design incompatibility challenges, we need significant re-work
> > on
> > > MXNet Module, KVStore, 

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Xingjian SHI]

-1. We do not need to get the eyeballs this way. Asking users to use features 
that are not mature enough will only hurt the future development of MXNet.


Xingjian


From: Naveen Swamy 
Sent: Saturday, March 24, 2018 2:54 AM
To: dev@mxnet.incubator.apache.org
Subject: Re: Request for comments - Keras-MXNet as submodule in MXNet

The proposal is about bringing a forked version of Keras(that works only
with MXNet) into Apache MXNet repo submodule that way MXNet gets more
eyeballs from existing Keras users and eventually Gluon, etc., , like
Sandeep mentioned Keras has a large user base which MXNet could tap into.

On Fri, Mar 23, 2018 at 11:50 AM, Yao Wang  wrote:

> -1 Creating Keras as submodule of MXNet will provide users a feeling that
> MXNet depends on Keras. Keras is a frontend library which can be supported
> by various different backend framework. It would be better to add backend
> framework as Keras's submodule(Keras depends on MXNet) rather than
> opposite.
>
> Best,
> Yao
>
> 2018-03-23 11:44 GMT-07:00 Xingjian SHI :
>
> > -1. I think we should wait until it's merged into keras-team/keras. The
> > repo is still not mature enough.
> >
> >
> > Best,
> >
> > Xingjian
> >
> >
> > 
> > From: sandeep krishnamurthy 
> > Sent: Friday, March 23, 2018 1:49 PM
> > To: dev@mxnet.incubator.apache.org
> > Subject: Request for comments - Keras-MXNet as submodule in MXNet
> >
> > Hello MXNet Community,
> >
> > Along with Lai, Karan and other MXNet contributors, I am working on
> adding
> > MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> > across CNNs and RNNs.
> > https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
[https://avatars0.githubusercontent.com/u/32447491?s=400=4]

deep-learning-tools/keras
github.com
keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or CNTK.



> > [https://avatars0.githubusercontent.com/u/32447491?s=400=4] > github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
> >
> > deep-learning-tools/keras > tools/keras/tree/keras2_mxnet_backend>
> > github.com
> > keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or
> > CNTK.
> >
> >
> >
> >
> > We wanted to gather the community feedback on the proposal for including
> > this keras-mxnet package as a submodule in Apache MXNet. This will enable
> > providing the Keras interface for MXNet users. MXNet users can choose
> Keras
> > interface for building their Neural Networks in Symbolic Mode (Ex:
> > mx.keras).
> >
> > *Advantages:*
> >
> > 1. Keras is widely popular interface that many DL practitioners are
> > familiar. By including keras interface within MXNet natively, we enable
> > many users to use MXNet with 0 learning curve.
> >
> > 2.  Adding as submodule and exposing natively within MXNet pip package,
> > would greatly enhance user experience and get more users as compared to
> > releasing a fork repository independently.
> >
> > 3. Why submodule? - Helps in easily managing with patching the latest
> > parent keras-team/keras developments and releases. Thereby helping us
> > provide users the core keras experience. Operational management.
> >
> > 4. Other minor advantages - Operational maintenance, pip, CI and quality
> > control.
> >
> > Please do share your comments on the proposal.
> >
> > Best,
> > Sandeep
> >
> > *Note: *We tried merging with keras-team/keras and we created a PR
> >  as well. However, due to
> > multiple design incompatibility challenges, we need significant re-work
> on
> > MXNet Module, KVStore, Optimizers to address keras-team design concerns.
> > Since, we are adhering to keras API interface exposed to users, we are
> > planning release on the forked repo for now. More details on the design
> > challenges and workaround tried -
> > https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-
> > VevdLrdn_eNd3nE4/edit?usp=sharing
> >
>

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Yuan Tang]

-1 I don't think this has significant difference. The adoption from
existing Keras users will only come once it's landed in Keras repo. I doubt
that many people will use it unless it's more stable and backward
compatibility is guaranteed.

On Fri, Mar 23, 2018 at 2:54 PM, Naveen Swamy  wrote:

> The proposal is about bringing a forked version of Keras(that works only
> with MXNet) into Apache MXNet repo submodule that way MXNet gets more
> eyeballs from existing Keras users and eventually Gluon, etc., , like
> Sandeep mentioned Keras has a large user base which MXNet could tap into.
>
> On Fri, Mar 23, 2018 at 11:50 AM, Yao Wang 
> wrote:
>
> > -1 Creating Keras as submodule of MXNet will provide users a feeling that
> > MXNet depends on Keras. Keras is a frontend library which can be
> supported
> > by various different backend framework. It would be better to add backend
> > framework as Keras's submodule(Keras depends on MXNet) rather than
> > opposite.
> >
> > Best,
> > Yao
> >
> > 2018-03-23 11:44 GMT-07:00 Xingjian SHI :
> >
> > > -1. I think we should wait until it's merged into keras-team/keras. The
> > > repo is still not mature enough.
> > >
> > >
> > > Best,
> > >
> > > Xingjian
> > >
> > >
> > > 
> > > From: sandeep krishnamurthy 
> > > Sent: Friday, March 23, 2018 1:49 PM
> > > To: dev@mxnet.incubator.apache.org
> > > Subject: Request for comments - Keras-MXNet as submodule in MXNet
> > >
> > > Hello MXNet Community,
> > >
> > > Along with Lai, Karan and other MXNet contributors, I am working on
> > adding
> > > MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> > > across CNNs and RNNs.
> > > https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
> > > [https://avatars0.githubusercontent.com/u/32447491?s=400=4] > > github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
> > >
> > > deep-learning-tools/keras > > tools/keras/tree/keras2_mxnet_backend>
> > > github.com
> > > keras - Deep Learning library for Python. Runs on TensorFlow, Theano,
> or
> > > CNTK.
> > >
> > >
> > >
> > >
> > > We wanted to gather the community feedback on the proposal for
> including
> > > this keras-mxnet package as a submodule in Apache MXNet. This will
> enable
> > > providing the Keras interface for MXNet users. MXNet users can choose
> > Keras
> > > interface for building their Neural Networks in Symbolic Mode (Ex:
> > > mx.keras).
> > >
> > > *Advantages:*
> > >
> > > 1. Keras is widely popular interface that many DL practitioners are
> > > familiar. By including keras interface within MXNet natively, we enable
> > > many users to use MXNet with 0 learning curve.
> > >
> > > 2.  Adding as submodule and exposing natively within MXNet pip package,
> > > would greatly enhance user experience and get more users as compared to
> > > releasing a fork repository independently.
> > >
> > > 3. Why submodule? - Helps in easily managing with patching the latest
> > > parent keras-team/keras developments and releases. Thereby helping us
> > > provide users the core keras experience. Operational management.
> > >
> > > 4. Other minor advantages - Operational maintenance, pip, CI and
> quality
> > > control.
> > >
> > > Please do share your comments on the proposal.
> > >
> > > Best,
> > > Sandeep
> > >
> > > *Note: *We tried merging with keras-team/keras and we created a PR
> > >  as well. However, due
> to
> > > multiple design incompatibility challenges, we need significant re-work
> > on
> > > MXNet Module, KVStore, Optimizers to address keras-team design
> concerns.
> > > Since, we are adhering to keras API interface exposed to users, we are
> > > planning release on the forked repo for now. More details on the design
> > > challenges and workaround tried -
> > > https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-
> > > VevdLrdn_eNd3nE4/edit?usp=sharing
> > >
> >
>

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Naveen Swamy]

The proposal is about bringing a forked version of Keras(that works only
with MXNet) into Apache MXNet repo submodule that way MXNet gets more
eyeballs from existing Keras users and eventually Gluon, etc., , like
Sandeep mentioned Keras has a large user base which MXNet could tap into.

On Fri, Mar 23, 2018 at 11:50 AM, Yao Wang  wrote:

> -1 Creating Keras as submodule of MXNet will provide users a feeling that
> MXNet depends on Keras. Keras is a frontend library which can be supported
> by various different backend framework. It would be better to add backend
> framework as Keras's submodule(Keras depends on MXNet) rather than
> opposite.
>
> Best,
> Yao
>
> 2018-03-23 11:44 GMT-07:00 Xingjian SHI :
>
> > -1. I think we should wait until it's merged into keras-team/keras. The
> > repo is still not mature enough.
> >
> >
> > Best,
> >
> > Xingjian
> >
> >
> > 
> > From: sandeep krishnamurthy 
> > Sent: Friday, March 23, 2018 1:49 PM
> > To: dev@mxnet.incubator.apache.org
> > Subject: Request for comments - Keras-MXNet as submodule in MXNet
> >
> > Hello MXNet Community,
> >
> > Along with Lai, Karan and other MXNet contributors, I am working on
> adding
> > MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> > across CNNs and RNNs.
> > https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
> > [https://avatars0.githubusercontent.com/u/32447491?s=400=4] > github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
> >
> > deep-learning-tools/keras > tools/keras/tree/keras2_mxnet_backend>
> > github.com
> > keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or
> > CNTK.
> >
> >
> >
> >
> > We wanted to gather the community feedback on the proposal for including
> > this keras-mxnet package as a submodule in Apache MXNet. This will enable
> > providing the Keras interface for MXNet users. MXNet users can choose
> Keras
> > interface for building their Neural Networks in Symbolic Mode (Ex:
> > mx.keras).
> >
> > *Advantages:*
> >
> > 1. Keras is widely popular interface that many DL practitioners are
> > familiar. By including keras interface within MXNet natively, we enable
> > many users to use MXNet with 0 learning curve.
> >
> > 2.  Adding as submodule and exposing natively within MXNet pip package,
> > would greatly enhance user experience and get more users as compared to
> > releasing a fork repository independently.
> >
> > 3. Why submodule? - Helps in easily managing with patching the latest
> > parent keras-team/keras developments and releases. Thereby helping us
> > provide users the core keras experience. Operational management.
> >
> > 4. Other minor advantages - Operational maintenance, pip, CI and quality
> > control.
> >
> > Please do share your comments on the proposal.
> >
> > Best,
> > Sandeep
> >
> > *Note: *We tried merging with keras-team/keras and we created a PR
> >  as well. However, due to
> > multiple design incompatibility challenges, we need significant re-work
> on
> > MXNet Module, KVStore, Optimizers to address keras-team design concerns.
> > Since, we are adhering to keras API interface exposed to users, we are
> > planning release on the forked repo for now. More details on the design
> > challenges and workaround tried -
> > https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-
> > VevdLrdn_eNd3nE4/edit?usp=sharing
> >
>

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Yao Wang]

-1 Creating Keras as submodule of MXNet will provide users a feeling that
MXNet depends on Keras. Keras is a frontend library which can be supported
by various different backend framework. It would be better to add backend
framework as Keras's submodule(Keras depends on MXNet) rather than opposite.

Best,
Yao

2018-03-23 11:44 GMT-07:00 Xingjian SHI :

> -1. I think we should wait until it's merged into keras-team/keras. The
> repo is still not mature enough.
>
>
> Best,
>
> Xingjian
>
>
> 
> From: sandeep krishnamurthy 
> Sent: Friday, March 23, 2018 1:49 PM
> To: dev@mxnet.incubator.apache.org
> Subject: Request for comments - Keras-MXNet as submodule in MXNet
>
> Hello MXNet Community,
>
> Along with Lai, Karan and other MXNet contributors, I am working on adding
> MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> across CNNs and RNNs.
> https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
> [https://avatars0.githubusercontent.com/u/32447491?s=400=4] github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend>
>
> deep-learning-tools/keras tools/keras/tree/keras2_mxnet_backend>
> github.com
> keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or
> CNTK.
>
>
>
>
> We wanted to gather the community feedback on the proposal for including
> this keras-mxnet package as a submodule in Apache MXNet. This will enable
> providing the Keras interface for MXNet users. MXNet users can choose Keras
> interface for building their Neural Networks in Symbolic Mode (Ex:
> mx.keras).
>
> *Advantages:*
>
> 1. Keras is widely popular interface that many DL practitioners are
> familiar. By including keras interface within MXNet natively, we enable
> many users to use MXNet with 0 learning curve.
>
> 2.  Adding as submodule and exposing natively within MXNet pip package,
> would greatly enhance user experience and get more users as compared to
> releasing a fork repository independently.
>
> 3. Why submodule? - Helps in easily managing with patching the latest
> parent keras-team/keras developments and releases. Thereby helping us
> provide users the core keras experience. Operational management.
>
> 4. Other minor advantages - Operational maintenance, pip, CI and quality
> control.
>
> Please do share your comments on the proposal.
>
> Best,
> Sandeep
>
> *Note: *We tried merging with keras-team/keras and we created a PR
>  as well. However, due to
> multiple design incompatibility challenges, we need significant re-work on
> MXNet Module, KVStore, Optimizers to address keras-team design concerns.
> Since, we are adhering to keras API interface exposed to users, we are
> planning release on the forked repo for now. More details on the design
> challenges and workaround tried -
> https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-
> VevdLrdn_eNd3nE4/edit?usp=sharing
>

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Zhi Zhang]

On 2018/03/23 05:49:07, sandeep krishnamurthy  
wrote: 
> Hello MXNet Community,
> 
> Along with Lai, Karan and other MXNet contributors, I am working on adding
> MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> across CNNs and RNNs.
> https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
> 
> We wanted to gather the community feedback on the proposal for including
> this keras-mxnet package as a submodule in Apache MXNet. This will enable
> providing the Keras interface for MXNet users. MXNet users can choose Keras
> interface for building their Neural Networks in Symbolic Mode (Ex:
> mx.keras).
> 
> *Advantages:*
> 
> 1. Keras is widely popular interface that many DL practitioners are
> familiar. By including keras interface within MXNet natively, we enable
> many users to use MXNet with 0 learning curve.
> 
> 2.  Adding as submodule and exposing natively within MXNet pip package,
> would greatly enhance user experience and get more users as compared to
> releasing a fork repository independently.
> 
> 3. Why submodule? - Helps in easily managing with patching the latest
> parent keras-team/keras developments and releases. Thereby helping us
> provide users the core keras experience. Operational management.
> 
> 4. Other minor advantages - Operational maintenance, pip, CI and quality
> control.
> 
> Please do share your comments on the proposal.
> 
> Best,
> Sandeep
> 
> *Note: *We tried merging with keras-team/keras and we created a PR
>  as well. However, due to
> multiple design incompatibility challenges, we need significant re-work on
> MXNet Module, KVStore, Optimizers to address keras-team design concerns.
> Since, we are adhering to keras API interface exposed to users, we are
> planning release on the forked repo for now. More details on the design
> challenges and workaround tried -
> https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-VevdLrdn_eNd3nE4/edit?usp=sharing
>  -1.
Submodules implies that they are to be used as dependencies. Apparently it is 
not the case. 

Re: MXNet C++ package improvements

[Eric Xie]

I see several issues with the design. I've commented in the document but for 
record here:

1. cpp-package is almost only used for inference. since you are planning a 
rewrite that's almost certainly non-backward-compatible, we might as well 
create a new interface that's inference only.

2. The hour-glass CAPI design should be kept.

On 2018/03/14 18:07:49, Anton Chernov  wrote: 
> Dear MxNet Community,
> 
> please find here
>  
> the
> design document for the proposed MXNet C++ package improvements for review
> and consideration.
> 
> Feedback is welcome and highly appreciated. Thank you!
> 
> BR
> Anton
> 

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Xingjian SHI]

-1. I think we should wait until it's merged into keras-team/keras. The repo is 
still not mature enough.


Best,

Xingjian



From: sandeep krishnamurthy 
Sent: Friday, March 23, 2018 1:49 PM
To: dev@mxnet.incubator.apache.org
Subject: Request for comments - Keras-MXNet as submodule in MXNet

Hello MXNet Community,

Along with Lai, Karan and other MXNet contributors, I am working on adding
MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
across CNNs and RNNs.
https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
[https://avatars0.githubusercontent.com/u/32447491?s=400=4]

deep-learning-tools/keras
github.com
keras - Deep Learning library for Python. Runs on TensorFlow, Theano, or CNTK.




We wanted to gather the community feedback on the proposal for including
this keras-mxnet package as a submodule in Apache MXNet. This will enable
providing the Keras interface for MXNet users. MXNet users can choose Keras
interface for building their Neural Networks in Symbolic Mode (Ex:
mx.keras).

*Advantages:*

1. Keras is widely popular interface that many DL practitioners are
familiar. By including keras interface within MXNet natively, we enable
many users to use MXNet with 0 learning curve.

2.  Adding as submodule and exposing natively within MXNet pip package,
would greatly enhance user experience and get more users as compared to
releasing a fork repository independently.

3. Why submodule? - Helps in easily managing with patching the latest
parent keras-team/keras developments and releases. Thereby helping us
provide users the core keras experience. Operational management.

4. Other minor advantages - Operational maintenance, pip, CI and quality
control.

Please do share your comments on the proposal.

Best,
Sandeep

*Note: *We tried merging with keras-team/keras and we created a PR
 as well. However, due to
multiple design incompatibility challenges, we need significant re-work on
MXNet Module, KVStore, Optimizers to address keras-team design concerns.
Since, we are adhering to keras API interface exposed to users, we are
planning release on the forked repo for now. More details on the design
challenges and workaround tried -
https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-VevdLrdn_eNd3nE4/edit?usp=sharing

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Eric Xie]

-1

If you make MXNet a submodule of keras, then you should PR that to keras.
If you want something like mxnet.keras, then you should do a full rewrite that 
only keeps the keras interface.

On 2018/03/23 05:49:07, sandeep krishnamurthy  
wrote: 
> Hello MXNet Community,
> 
> Along with Lai, Karan and other MXNet contributors, I am working on adding
> MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> across CNNs and RNNs.
> https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
> 
> We wanted to gather the community feedback on the proposal for including
> this keras-mxnet package as a submodule in Apache MXNet. This will enable
> providing the Keras interface for MXNet users. MXNet users can choose Keras
> interface for building their Neural Networks in Symbolic Mode (Ex:
> mx.keras).
> 
> *Advantages:*
> 
> 1. Keras is widely popular interface that many DL practitioners are
> familiar. By including keras interface within MXNet natively, we enable
> many users to use MXNet with 0 learning curve.
> 
> 2.  Adding as submodule and exposing natively within MXNet pip package,
> would greatly enhance user experience and get more users as compared to
> releasing a fork repository independently.
> 
> 3. Why submodule? - Helps in easily managing with patching the latest
> parent keras-team/keras developments and releases. Thereby helping us
> provide users the core keras experience. Operational management.
> 
> 4. Other minor advantages - Operational maintenance, pip, CI and quality
> control.
> 
> Please do share your comments on the proposal.
> 
> Best,
> Sandeep
> 
> *Note: *We tried merging with keras-team/keras and we created a PR
>  as well. However, due to
> multiple design incompatibility challenges, we need significant re-work on
> MXNet Module, KVStore, Optimizers to address keras-team design concerns.
> Since, we are adhering to keras API interface exposed to users, we are
> planning release on the forked repo for now. More details on the design
> challenges and workaround tried -
> https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-VevdLrdn_eNd3nE4/edit?usp=sharing
> 

RE: call for contributions to next MXNet release

[Zhao, Patric]

Really thank Marco, Da and other reviewers' help :)  

I'd like to update the status of MKL-DNN bugs. 
Feel free to let me know if there're any other issues.

There're 8 opened issues (1 discussion thread is not included).

2 issues are WIP and will be completed in next several days.
#10189, Race condition when MKLDNN is enabled MKL, Zheng Da  
Note: this data race exist in theory but couldn't happen now. Da is working on 
the enhancement to make it perfect.
-#8712, test_operator.test_depthwise_convolution fails in [Python2: MKLML-CPU] 
and [Python3: MKLML-CPU], Patric Zhao

6 opened issues have been resolved and wait to review/merge/close
#10168 (PR#10218) CPP test case fails to compile using gcc 4.8.5 when MKLDNN 
enabled Breaking Build MKL
#10141 (Passed, to be closed)  Flaky test 
test_operator_gpu.test_convolution_options @ Python2: MKLDNN-GPU Flaky MKL Test
#10092 (PR#10021) No "full" pooling convention support with MKL-DNN Feature 
request MKL Operator
#10026 (PR#10069) MXNET_MKLDNN_DEBUG=1 produces errors Bug MKL
#8974 (Question, will update doc) MXNET compatibility with MKL libraries 
bundled in Microsoft R Open MKL
#8532 (resolved by workaround) mxnet-mkl (v0.12.0) crash when using 
(conda-installed) numpy with MKL Bug MKL
Note: The duplicated libiomp.so library caused the conflict in the conda and 
it's not introduced by MKL-DNN. 
It can be resolved by environment variable. I am looking for C API 
to set in the runtime.

> -Original Message-
> From: Marco de Abreu [mailto:marco.g.ab...@googlemail.com]
> Sent: Saturday, March 17, 2018 1:37 AM
> To: dev@mxnet.incubator.apache.org
> Subject: Re: call for contributions to next MXNet release
> 
> Hi Patric,
> 
> I have added three more issues:
> https://github.com/apache/incubator-mxnet/issues/10131
> https://github.com/apache/incubator-mxnet/issues/10133
> https://github.com/apache/incubator-mxnet/issues/10134
> 
> Also, you missed https://github.com/apache/incubator-mxnet/issues/8712 -
> sorry, it was not labelled before
> 
> Best regards,
> Marco
> 
> On Fri, Mar 16, 2018 at 7:45 AM, Zhao, Patric 
> wrote:
> 
> > MKL issues summary:
> >
> > Feel free to let me know if anything I missed.
> >
> > Totally, there’re 11 open issues  in the github with the label of MKL
> > as
> > below:
> >  4 Bugs issues: #10092, #10086,  #10026, #8881
> >  4 Building issues: #9993, #9828, #8974, #8532
> >  1 Flaky Tests: #9864
> >  2 wrong label and invalidated:  #9844, #8874
> >
> > Current status.
> > 3 DONE, to be closed.
> > 4 WIP (working in progress)
> > 2 TODO
> > 2 are invalidated now.
> >
> > Details:
> > #10092, WIP, PR#10021
> > #10086, TODO
> > #10026, WIP
> > #9993, DONE (need to be closed), fixed by #10075 #9864, TODO #9844,
> > INVALIDED, not related w/ MKL #9828, DONE (need to be closed), fixed
> > by #9918 & #10115 #8974, WIP #8881, DONE (need to be closed), fixed by
> > #9112 #8874, INVALIDED, “I believe, we are not using MKL”
> > #8532, WIP, library conflict and can be resolved by environment
> > setting
> >
> > Lists:
> > No "full" pooling convention support with MKL-DNN Feature request MKL
> > Operator
> > #10092 opened 2 days ago by marcoabreu
> > https://github.com/apache/incubator-mxnet/issues/10092
> >
> > [MXNET-84] Segfault test_autograd.test_unary_func @ Python3:
> > MKLDNN-CPU Bug MKL Test
> > #10086 opened 3 days ago by marcoabreu
> > https://github.com/apache/incubator-mxnet/issues/10086
> >
> > MXNET_MKLDNN_DEBUG=1 produces errors Bug MKL
> > #10026 opened 8 days ago by marcoabreu
> > https://github.com/apache/incubator-mxnet/issues/10026
> >
> > cmake cannot build mxnet Bug Build MKL
> > #9993 opened 11 days ago by jacky4323
> > https://github.com/apache/incubator-mxnet/issues/9993
> >
> > Flaky hanging test_operator.test_laop_3 @ Python3: MKLDNN-CPU Flaky
> > MKL Test
> > #9864 opened 21 days ago by marcoabreu
> > https://github.com/apache/incubator-mxnet/issues/9864
> >
> > Flaky test_operator_gpu.test_binary_op @ Python3: MKLDNN-GPU Flaky
> MKL
> > Test
> > #9844 opened 23 days ago by marcoabreu
> > https://github.com/apache/incubator-mxnet/issues/9844
> >
> > Building with MKL fails on OSX Build MKL
> > #9828 opened 25 days ago by sbodenstein
> > https://github.com/apache/incubator-mxnet/issues/9828
> >
> > MXNET compatibility with MKL libraries bundled in Microsoft R Open MKL
> > #8974 opened on Dec 7, 2017 by mjmg
> > https://github.com/apache/incubator-mxnet/issues/8974
> >
> > Intel MKL FATAL ERROR: Cannot load libmkl_avx512_mic.so or
> libmkl_def.so.
> > Bug Data-loading MKL
> > #8881 opened on Nov 30, 2017 by wuzhijiexia
> > https://github.com/apache/incubator-mxnet/issues/8881
> >
> > mxnet installation from source: C++ linkage error on HPC C++
> > Installation MKL
> > #8874 opened on Nov 30, 2017 by jerrin92
> > https://github.com/apache/incubator-mxnet/issues/8874
> >
> > mxnet-mkl (v0.12.0) crash when using (conda-installed) numpy with MKL
> > Bug 

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Naveen Swamy]

+1

> On Mar 22, 2018, at 11:11 PM, Chris Olivier  wrote:
> 
> +1
> 
> On Thu, Mar 22, 2018 at 10:49 PM sandeep krishnamurthy <
> sandeep.krishn...@gmail.com> wrote:
> 
>> Hello MXNet Community,
>> 
>> Along with Lai, Karan and other MXNet contributors, I am working on adding
>> MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
>> across CNNs and RNNs.
>> https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
>> 
>> We wanted to gather the community feedback on the proposal for including
>> this keras-mxnet package as a submodule in Apache MXNet. This will enable
>> providing the Keras interface for MXNet users. MXNet users can choose Keras
>> interface for building their Neural Networks in Symbolic Mode (Ex:
>> mx.keras).
>> 
>> *Advantages:*
>> 
>> 1. Keras is widely popular interface that many DL practitioners are
>> familiar. By including keras interface within MXNet natively, we enable
>> many users to use MXNet with 0 learning curve.
>> 
>> 2.  Adding as submodule and exposing natively within MXNet pip package,
>> would greatly enhance user experience and get more users as compared to
>> releasing a fork repository independently.
>> 
>> 3. Why submodule? - Helps in easily managing with patching the latest
>> parent keras-team/keras developments and releases. Thereby helping us
>> provide users the core keras experience. Operational management.
>> 
>> 4. Other minor advantages - Operational maintenance, pip, CI and quality
>> control.
>> 
>> Please do share your comments on the proposal.
>> 
>> Best,
>> Sandeep
>> 
>> *Note: *We tried merging with keras-team/keras and we created a PR
>>  as well. However, due to
>> multiple design incompatibility challenges, we need significant re-work on
>> MXNet Module, KVStore, Optimizers to address keras-team design concerns.
>> Since, we are adhering to keras API interface exposed to users, we are
>> planning release on the forked repo for now. More details on the design
>> challenges and workaround tried -
>> 
>> https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-VevdLrdn_eNd3nE4/edit?usp=sharing
>> 

Re: Request for comments - Keras-MXNet as submodule in MXNet

[Chris Olivier]

+1

On Thu, Mar 22, 2018 at 10:49 PM sandeep krishnamurthy <
sandeep.krishn...@gmail.com> wrote:

> Hello MXNet Community,
>
> Along with Lai, Karan and other MXNet contributors, I am working on adding
> MXNet backend for Keras. Currently supporting around ~70% of Keras APIs
> across CNNs and RNNs.
> https://github.com/deep-learning-tools/keras/tree/keras2_mxnet_backend
>
> We wanted to gather the community feedback on the proposal for including
> this keras-mxnet package as a submodule in Apache MXNet. This will enable
> providing the Keras interface for MXNet users. MXNet users can choose Keras
> interface for building their Neural Networks in Symbolic Mode (Ex:
> mx.keras).
>
> *Advantages:*
>
> 1. Keras is widely popular interface that many DL practitioners are
> familiar. By including keras interface within MXNet natively, we enable
> many users to use MXNet with 0 learning curve.
>
> 2.  Adding as submodule and exposing natively within MXNet pip package,
> would greatly enhance user experience and get more users as compared to
> releasing a fork repository independently.
>
> 3. Why submodule? - Helps in easily managing with patching the latest
> parent keras-team/keras developments and releases. Thereby helping us
> provide users the core keras experience. Operational management.
>
> 4. Other minor advantages - Operational maintenance, pip, CI and quality
> control.
>
> Please do share your comments on the proposal.
>
> Best,
> Sandeep
>
> *Note: *We tried merging with keras-team/keras and we created a PR
>  as well. However, due to
> multiple design incompatibility challenges, we need significant re-work on
> MXNet Module, KVStore, Optimizers to address keras-team design concerns.
> Since, we are adhering to keras API interface exposed to users, we are
> planning release on the forked repo for now. More details on the design
> challenges and workaround tried -
>
> https://docs.google.com/document/d/1Vn5ip5MzCKcN29KCCnwjB2d59y-VevdLrdn_eNd3nE4/edit?usp=sharing
>